def testCopyHostToDevicePerformance(self, bits):
# do cuda arithmetic
f_gpu = GF2nStub.GF2nStub("Cuda", bits, -1)
a_gpu = f_gpu()
b_gpu = f_gpu()
GF2nStub.run("measureKernelLaunchOverhead", a_gpu, b_gpu)
metrics = GF2nStub.getMetrics("bn_a")
PerformanceDataLogger().addPerfResult("copy host -> device", bits,
"Cuda",
metrics["copyToDevice_time"])
def testCopyDeviceToHostPerformance(self, bits):
# do cuda arithmetic
f_gpu = GF2nStub.GF2nStub("Cuda", bits, -1)
a_gpu = f_gpu()
b_gpu = f_gpu()
res = GF2nStub.run("parAdd", a_gpu, b_gpu)
res_value = res._value
metrics = GF2nStub.getMetrics("res")
PerformanceDataLogger().addPerfResult("copy device -> host", bits,
"Cuda",
metrics["copyToHost_time"])